- The scheduler now has a per-thread current address, which prevents
  the naïve (exec → dequeue → next thread → next addr) schedule from
  causing a deadlock by never selecting an address used by the current
  thread
- Popping a frame now frees that frame's stack memory